Investigating a Cloudflare Workers AI failure
Check model-specific failures, response errors and the scope of a Workers AI incident.
2 min read
The CheckUpstream blog
Engineering,
upstream.
What to check when a vendor fails. How to map your dependencies. And what the evidence can actually tell you.
Browse all articlesStart here
What a vendor status page can tell you
Use vendor incident reports alongside endpoint measurements and your own request data. Each source answers a different part of the investigation.
1 min read
Three sources. Different questions.
- Vendor report
- What has the vendor acknowledged?
- Endpoint check
- Did this endpoint respond to a probe?
- Your requests
- Did the operation in your app succeed?
All articles
1 to 12 of 13 articles
Why CheckUpstream starts with your dependencies
Map the services in your code, connect vendor incidents to relevant projects, and review the evidence before acting or preparing a claim.
1 min read
Investigating API rate limits without creating a retry storm
Check rate limits, concurrent jobs and retry behaviour when API requests start failing.
1 min read
What a vendor status page can tell you
Use vendor incident reports alongside endpoint measurements and your own request data. Each source answers a different part of the investigation.
1 min read
Investigating Stripe issues across payments, webhooks, and checkout
Separate API errors, webhook delays and checkout failures in a Stripe investigation.
1 min read
What to check first when checkout starts failing
A short checkout incident runbook: establish the failing step, check internal and vendor changes, protect payment state, and record what happened.
1 min read
Deciding when a vendor incident should page someone
Write a response rule for each dependency: which user journeys it affects, how to confirm impact, and when a person needs to act.
1 min read
Alert on impact, not existence
Connect vendor incident reports to relevant projects, then use request evidence to understand application impact.
3 min read
How we count front door checks
Which checks count, what a count of reachable checks means, why we publish no percentage, and why vendor incident reports stay a separate record.
3 min read
What an SLA credit is actually worth
Calculate a conditional EC2 service credit, distinguish eligible spend from incident costs, and check the actual evidence and filing requirements.
2 min read
How a dependency failure can spread through your app
An illustrative failure chain: slow requests, repeated retries and a growing queue. What to inspect before treating every symptom as a separate outage.
2 min read
Finding vendor connections in your dependency files
Use dependency files to find possible vendor connections, then review which services your application actually calls.
3 min read